Download Stochastic Model of the Potential Spread of Highly Pathogenic Avian

Stochastic Model of the Potential Spread of Highly Pathogenic Avian Influenza
(HPAI) from an Infected Commercial Broiler Operation in Georgia, U.S.
DOREA, F.C1; VIEIRA, A.R.1,2; HOFACRE, C1; COLE, D.J.3
1-Poultry Diagnostic Research Center, UGA. 2- National Food Institute, Technical University of Denmark. 3-College of Public Health, UGA.
[email protected]
Summary
The potential spread of HPAI between commercial broiler farms in Georgia, U.S. was mathematically modeled. The dynamics of spread within the index flock was modeled using an SEIR
deterministic model, and predicted that grower detection of flock infection is most likely 5 days after virus introduction. Off farm spread of virus was estimated stochastically for this period,
predicting a range of exposed farms from 0-5 depending upon the density of farms in the area. Modeled off-farm spread was most frequently associated with feed trucks (highest frequency
of visits and range of farms visited) and company personnel or hired help (highest level of contact).
ON-FARM Spread Model – DETECTION DAY under diverse virus virulence characteristics
Mortality thresholds have been shown by other studies to be the most reliable indicator of AI infection in many poultry
production systems (1, 2). Our model used the average mortality threshold reported by Vieira et al (3) of 0.2% (reported
range 0.04 to 1.08%) to determine the most likely day of grower detection of flock infection on the index farm.
According to our model the most likely day of illness detection is the 5th day across the range of reported average
mortality levels tolerated by commercial poultry growers.
A compartmental epidemic (SEIR) model was used to estimate the days to grower detection of
flock infection following the introduction of a highly pathogenic avian influenza (HPAI) virus.
tx
tx+1
The estimated detection day of t=5 was fairly robust across varying parameters of:
Virus incubation period (range: 1 to 4 days)
Infectious period (range: 1 to 10 days) and
Transmission rates (22 to 140 birds per infected contact)
Dynamics of the SEIR model in a poultry house with a 21,000 bird capacity and introduction of a single
infected bird on day zero:
Infectious period = 6 days (incubation =2d)
Infectious period = 3 days
Infectious period = 6 days
25000
25000
20000
20000
1000
Infectious period = 3 days (incubation =2d)
t=3days
800
Number of DEAD birds
Number of DEAD birds
•Incubation period (σ) = 2 days
•Daily transmission rate (β) = 70 birds/infected bird
t=4days
600
t=5days
400
200
0
10
22
33
40
50
70
90
110 130 140
1000
600
400
200
0
10
10000
15000
10000
5000
5000
0
0
1
2
3
4
5
6
7
8
9
1
10 11 12 13 14 15 16 17
2
3
4
5
6
7
8
9
10
11
12 13
14
15
16
17
8
8
7
7
6
5
4
3
10
Using these parameters, time to detection was estimated for different scenarios of virus virulence, as shown in
the panel to the right.
33
40
50
70
90
110 130 140
6
5
4
3
Days
Days
22
Infectious period = 3 days (incubation =2d)
Detection day
Susceptible
Exposed
Infectious
Dead
Detection day
# of Birds
# of Birds
Infectious period = 6 days (incubation =2d)
15000
t=3days
t=4days
t=5days
800
22
33
40
50
70
90
110 130
140
10
Daily transmission rate (number of birds infected by a single bird)
0-7 days old
7-28 days olds
22
33
40
50
70
90
110 130
140
Daily transmission rate (number of birds infected by a single bird)
28 days to end
0-7 days old
7-28 days old
28 days to end
OFF-Farm Spread Model – translating contact information and farm density into predictions of HPAI spread among farms
Visitor comes to Index
farm
Bernoulli (p)
X
Visitor is contaminated
Bernoulli (p)
f(y)
Number of farms exposed
(Each visitor was assigned one of the
following distributions):
Uniform (min,max)
Poisson (λ)
β-Pert (min,most lik.,max)
If f(y) > 0,
the
visitor
spreads
p = probability of contamination
Uniform (min,max)
The stochastic off-farm model predicted the contact rate between farms, based upon the movement of people, vehicles, and equipment
from the index farm to other farms in the same day. The identification of human activities and their frequency were obtained from a poultry
grower survey. (3) Spread from the index farm was assumed to be possible after two days of infection, after the initially infected birds on the
reached the end of their incubation period and became infectious.
Spread is assumed to occur if:
(1) A visitor comes to the index farm. This event is simulated by a Bernoulli trial which probability of success is represented by the daily
chance of that specific visitor coming to the farm (flock frequency/49; as an average growout period would be of 49 days).
(2) The visitor contacts infected birds, or potentially infective material, and becomes contaminated. In the table, period refers to
whether the activity is performed during the interval between flocks (I), while flocks are in the houses (F), or possible at anytime (A). The
probability of a visitor getting contaminated (”Cont.”), for each activity, was classified as “none”, “very low”, “”low”, “moderate”, “high” and
“very high”. Each of the five non-zero risk categories is represented in the model by a uniform continuous distribution, with mutually exclusive
ranges of equal size (0.20) from zero to one. Once a probability is drawn in one trial, it’s used in a Bernoulli trial (yes-no event).
Prevalence is the proportion of specific scenarios considered for each activity, or the proportional weight of each listed item within the
same visitor type
(3) The visitor (now a vector) goes to other farms. The outcomes of the first two events are multiplied, and if the result is not zero (both
of them occur), the number of contacted farms will be determined by the range of visits for the vector. A uniform (discrete) distribution
represented the range in the number of farms that could be visited in a day after a visit to the index farm.
Table notes:
**Frequency of total house clean out, times the reported frequency of growers disposing litter off-farm
***No growers reported disposal of dead birds off-farm
Activity
Period
1 Clean Out Services
Owned Equipment
Borrowed Equipment
Contracted Services
2 Decaking, Housekeeping
Owned Equipment
Borrowed Equipment/Contracted
3 Disposal of Litter off-farm
4 Disposal of Birds off-farm
5 Live Haul
6 Shaving Suppliers
7 Repairmen
Growout / Inside
Growout / Outside
Downtime
8 LP Delivery
Winter / Downtime
Winter / Growout
Summer / Downtime
Summer / Growout
9 Feed Trucks
10 Chick Bus
Service Persons, Managers,
11 Vets
Utilities, Meter readers, Pit
12 inspect.
Downtime
Growout
13 Unauthorized Visitors
Inside
Outside
14 Vaccination Crews
Vaccine
No Vaccine
15 Grower, Hired Help
I
HIGH DENSITY COUNTY
Flock
Max farm
Freq.
Prev.~
range/day
0.33
0.50
0
0.10
2
0.40
2
0.25
0.90
0
0.10
2
0.25**
1.00
1
***
0
1.00
1
1.00
3
0.33
1.00
2
0.25
3
0.75
0.15
0.10
5
3.00
0.17
0.34
0.33
0.16
0.33
15.00
1.00
5
1.00
1.00
5
5
7.00
1.00
15
1.00
0.25
0.75
4.00
2
0.25
0.75
0.50
5
0.50
0.50
14.00
1.00
2
Cont.
0
Low
Mod.
I
I
F
F
I
A
0
Low
VH
VH
VH
Mod.
High
Mod.
Low
A
F
I
Low
Mod.
Low
Mod.
Mod.
Mod.
F
VH
A
Low
Mod.
A
VH
Low
F
F
High
0
Mod.
LOW DENSITY COUNTY
Flock
Max farm
Freq.
Prev.~
range/day
0.33
0.90
0
0.00
2
0.10
2
0.25
0.95
0
0.05
2
0.25**
1.00
2
***
0
1.00
1
1.00
1
0.33
1.00
2
0.25
3
0.80
0.10
0.10
3
3.00
0.17
0.34
0.55
0.16
0.33
15.00
1.00
5
1.00
1.00
5
4
7.00
1.00
10
1.00
0.25
0.75
2.00
2
0.50
0.50
0.50
5
0.50
0.50
14.00
1.00
2
Results
Number of farms potentially exposed to contamination from an infected index farm
via human movement in a 24hr interval in a high poultry farm density region (1.45
farms/5 mi2) and a low poultry farm density region (0.49 farms/5 mi2)
High farm density county
Proportional contribution of each listed visitor to the
overall spread of HPAI virus from one infected farm
in one day
40%
35%
Low farm density county
30%
2.5
2
1.5
Maximum
Mean
1
Minimum
0.5
0
2.5
2
1.5
Maximum
Mean
1
Probability of a HPAI infection being restricted to the index farm
Farm density
Day 0
Day 1
Day 2*
Day 3
Day 4
Day 5
Low
100%
100%
69.5%
47.6%
33.3%
22.3%
High
100%
100%
67.4%
47.5%
33.3%
23.1%
Day 6
15.3%
16.3%
Percentage
3
Number of exposed farms (LOG10)
3
Number of exposed farms (LOG10)
Probability that HPAI infection is not transmitted from the index farm in the days
following initial flock infection (Day 0). Calculations assume that off-farm spread of
virus from the index farm is not possible until Day 2.
25%
20%
15%
10%
5%
0%
Minimum
0.5
0
0
1
2
3
4
5
6
0
1
Days
2
3
4
5
6
Visitor/Farm activity
Days
Low farm density
High farm density
Conclusions and discussion
Under several potential scenarios, an introduction of HPAI in commercial poultry farms would be detected in the 5th day of spread.
Immediate reporting following detection is critical to prevent expansion of outbreak, since the number of exposed farms increases exponentially by the time of detection on the index farm.
Most off-farm exposures were associated with feed trucks and human contacts associated with in-house contact with birds, such as company service personnel and hired help. Use of non-family hired help was reportedly
more common in the low farm density area, while the growers in the highly poultry dense area were more likely to report visits from non-company personnel (unauthorized visitors), which is reflected in the results.
The actual number of infected farms resulting from exposure to a single index farm would be lower than the estimates shown here. The reduction of transmission risk associated with on-farm biosecurity and virus decay
during transport were not included in this model. Nonetheless, by using data on detection thresholds and horizontal contact rates reported by broiler poultry growers in Georgia, we were able to model the likely
pathways for virus spread under realistic conditions.
The observed results highlight the need for early reporting of flock illness and the importance of biosecurity focused on key high risk activities.
NEXT STEPS : Role of biosecurity measures – modeling the effectiveness in reducing HPAI spread
This work showed the potential risk of farm exposure to a highly virulent virus associated with human movement between an infected commercial poultry farm and other susceptible farms. This made clear that further
developments of the model are needed to examine the role of farm biosecurity in preventing virus spread. To expand the model, two extra sources of information will be used:
(1)Field experiments demonstrating the efficiency of specific biosecurity measures;
(2)Information regarding the biosecurity measures as applied by broiler growers in Georgia. This information was obtained via a survey of growers.
Results of the survey of growers:
Which biosecurity measures are routinely adopted to reduce the
chance of a visitor leaving the farm contaminated.
Visitor comes to Index
farm
Bernoulli (p)
X
Visitor is contaminated
Bernoulli (p)
p = probability of contamination
Uniform (min,max)
X
Model expansion:
Visitor leaves the farm
contaminated
(person or vehicle)
Bernoulli (p)
Field experiments
How effective each of the practiced biosecurity measures
is on preventing contamination of the visitor
f(y)
If f(y) > 0,
the
visitor
spreads
References:
(1) Elbers, A. R., Holtslag, J. B., Bouma, A., and Koch, G. Within-flock mortality during the high-pathogenicity avian influenza (H7N7) epidemic in The Netherlands in 2003: implications for an
early detection system. Avian Dis. 51:304-308. 2007.
(2)Savill, N. J., St Rose, S. G., and Woolhouse, M. E. Detection of mortality clusters associated with highly pathogenic avian influenza in poultry: a theoretical analysis. J. R. Soc. Interface . 2008
(3)Vieira, A. R., Hofacre, C. L., Smith, J. A., and Cole, D. Human contacts and potential pathways of disease introduction on Georgia poultry farms. Avian Diseases . In press
Number of farms exposed
(Each visitor was assigned one of the
following distributions):
Uniform (min,max)
Poisson (λ)
β-Pert (min,most lik.,max)
Acknowledgements:
Project funding provided by a cooperative USDA grant
Participating Commercial Poultry Companies